Commentary generation for financial markets

Financial markets are based on the daily movements of thousands of tradable assets, such as stocks, resulting in billion-dollar trade volumes and affecting investors and companies around the globe. In this volatile and high-stakes environment, financial-service firms employ analysts to create compact market commentaries that serve as insightful summaries with key pieces of information. In this work, we attempt to automate this process by formally defining and algorithmically solving the Market Commentary Generation (MCG) problem. In addition to saving time and cost via automation, our approach makes a number of contributions that differentiate it from previous related work. These include the consideration of thousands of underlying time series, the ability to capture and encode significant market events that involve multiple financial entities, and the ability to deliver high quality commentary even in the presence of small and unlabeled historical datasets. Finally, our approach takes into account the strict compliance requirements of the finance domain, which prevent the use of black-box methods that can produce language that violates key rules and regulations. We compare our work against competitive baselines via an evaluation that includes both qualitative and quantitative experiments.